x64: Fix f64x2.splat without SSE3 (#13011)

Don't sink a load into `pshufd` which loads 16 bytes, instead force
`put_in_xmm` to ensure only 8 bytes are loaded.

Author: alexcrichton

cranelift/codegen/src/isa/x64/lower.isle

@@ -4903,7 +4903,7 @@
 (rule 0 (lower (has_type $I64X2 (splat _ src)))
         (x64_pshufd (bitcast_gpr_to_xmm 64 src) 0b01_00_01_00))
 (rule 0 (lower (has_type $F64X2 (splat _ src)))
-        (x64_pshufd src 0b01_00_01_00))
+        (x64_pshufd (put_in_xmm src) 0b01_00_01_00))
 (rule 6 (lower (has_type (multi_lane 64 2) (splat _ (sinkable_load addr))))
         (if-let true (has_sse3))
         (x64_movddup addr))

tests/misc_testsuite/simd/edge-of-memory.wast

@@ -336,3 +336,10 @@
 (assert_return (invoke "2-byte-from-end" (i32.const 65534)))
 (assert_return (invoke "4-byte-from-end" (i32.const 65532)))
 (assert_return (invoke "8-byte-from-end" (i32.const 65528)))
+
+(module
+  (memory 1)
+  (func (export "test") (param i32) (result v128)
+    (f64x2.splat (f64.load (local.get 0)))))
+
+(assert_return (invoke "test" (i32.const 65528)) (v128.const f64x2 0 0))